DeepVOX: Discovering Features from Raw Audio for Speaker Recognition in Non-ideal Audio Signals

Automatic speaker recognition algorithms typically use pre-defined filterbanks, such as Mel-Frequency and Gammatone filterbanks, for characterizing speech audio. However, it has been observed that the features extracted using these filterbanks are not resilient to diverse audio degradations. In this work, we propose a deep learning-based technique to deduce the filterbank design from vast amounts of speech audio. The purpose of such a filterbank is to extract features robust to non-ideal audio conditions, such as degraded, short duration, and multi-lingual speech. To this effect, a 1D convolutional neural network is designed to learn a time-domain filterbank called DeepVOX directly from raw speech audio. Secondly, an adaptive triplet mining technique is developed to efficiently mine the data samples best suited to train the filterbank. Thirdly, a detailed ablation study of the DeepVOX filterbanks reveals the presence of both vocal source and vocal tract characteristics in the extracted features. Experimental results on VOXCeleb2, NIST SRE 2008, 2010 and 2018, and Fisher speech datasets demonstrate the efficacy of the DeepVOX features across a variety of degraded, short duration, and multi-lingual speech. The DeepVOX features also shown to improve the performance of existing speaker recognition algorithms, such as the xVector-PLDA and the iVector-PLDA.